两种面向推荐系统的数据压缩方法

今日头条的服务器每天都会产生规模庞大的训练数据,为方便进行训练,这些数据都具有特定的格式和分布特征。使用不同类型的通用压缩算法(字典类型及非字典类型)进行测试,发现单独任何一种算法都无法在满足业务需求(速率需求和CPU占比等)的同时获得较为可观的压缩比。针对今日头条的训练数据,提出了分段聚类压缩和Hash recoding压缩两种策略。实验结果表明,分段聚类压缩方式在更好地保证了压缩率的同时提高了压缩速率的目的;Hash recoding压缩方式则更好地达到了以少量压缩速率的损失换取更可观的压缩率的目的。分段聚类方式搭配Gzip压缩算法的组合能使压缩速率提高300%以上;Hash recoding方式匹配Snappy压缩算法能使压缩率缩小50%以上。根据实际需求,不论选择哪种策略,对于降低今日头条的运营成本,提升业务处理的效率,提供更好的用户体验,都有一定意义。     

Combination of bottleneck feature extraction and dereverberation for distant-talking speech recognition

The performance of speech recognition in distant-talking environments is severely degraded by the reverberation that can occur in enclosed spaces (e.g., meeting rooms). To mitigate this degradation, dereverberation techniques such as network structure-based denoising autoencoders and multi-step linear prediction are used to improve the recognition accuracy of reverberant speech. Regardless of the reverberant conditions, a novel discriminative bottleneck feature extraction approach has been demonstrated to be effective for speech recognition under a range of conditions. As bottleneck feature extraction is not primarily designed for dereverberation, we are interested in whether it can compensate for other carefully designed dereverberation approaches. In this paper, we propose three schemes covering both front-end processing (cascaded combination and parallel combination) and back-end processing (system combination). Each of these schemes integrates bottleneck feature extraction with dereverberation. The effectiveness of these schemes is evaluated via a series of experiments using the REVERB challenge dataset.     

A deep bidirectional LSTM approach for video-realistic talking head

This paper proposes a deep bidirectional long short-term memory approach in modeling the long contextual, nonlinear mapping between audio and visual streams for video-realistic talking head. In training stage, an audio-visual stereo database is firstly recorded as a subject talking to a camera. The audio streams are converted into acoustic feature, i.e. Mel-Frequency Cepstrum Coefficients (MFCCs), and their textual labels are also extracted. The visual streams, in particular, the lower face region, are compactly represented by active appearance model (AAM) parameters by which the shape and texture variations can be jointly modeled. Given pairs of the audio and visual parameter sequence, a DBLSTM model is trained to learn the sequence mapping from audio to visual space. For any unseen speech audio, whether it is original recorded or synthesized by text-to-speech (TTS), the trained DBLSTM model can predict a convincing AAM parameter trajectory for the lower face animation. To further improve the realism of the proposed talking head, the trajectory tiling method is adopted to use the DBLSTM predicted AAM trajectory as a guide to select a smooth real sample image sequence from the recorded database. We then stitch the selected lower face image sequence back to a background face video of the same subject, resulting in a video-realistic talking head. Experimental results show that the proposed DBLSTM approach outperforms the existing HMM-based approach in both objective and subjective evaluations.     

Practical chosen-ciphertext secure Hierarchical Identity-Based Broadcast Encryption

We focus on practical Hierarchical Identity-Based Broadcast Encryption (HIBBE) with semantic security against adaptively chosen-ciphertext attacks (CCA2) in the standard model. We achieve this goal in two steps. First, we propose a new HIBBE scheme that is secure against chosen-plaintext attacks (CPA). Compared with the existing HIBBE scheme that is built from composite-order bilinear groups, our construction is based on prime-order bilinear groups. The much better efficiency of group operations in prime-order bilinear groups makes our proposed HIBBE scheme more practical. Then, we convert it into a CCA2-secure scheme at the cost of a one-time signature. Instead of extending one user hierarchy in the Canetti–Halevi–Katz approach from CPA-secure (\(l+1\))-Hierarchical Identity-Based Encryption [(\(l+1\))-HIBE] to CCA2-secure \(l\)-HIBE, our construction merely adds one on-the-fly dummy user in the basic scheme. We formally prove the security of these two schemes in the standard model. Comprehensive theoretical analyses and experimental results demonstrate that the proposed HIBBE schemes achieve desirable performance.     

An efficient algorithm for partially matched services in internet of services

Internet of Things (IoT) connects billions of devices in an Internet-like structure. Each device encapsulated as a real-world service which provides functionality and exchanges information with other devices. This large-scale information exchange results in new interactions between things and people. Unlike traditional web services, internet of services is highly dynamic and continuously changing due to constant degrade, vanish and possibly reappear of the devices, this opens a new challenge in the process of resource discovery and selection. In response to increasing numbers of services in the discovery and selection process, there is a corresponding increase in number of service consumers and consequent diversity of quality of service (QoS) available. Increase in both sides’ leads to the diversity in the demand and supply of services, which would result in the partial match of the requirements and offers. This paper proposed an IoT service ranking and selection algorithm by considering multiple QoS requirements and allowing partially matched services to be counted as a candidate for the selection process. One of the applications of IoT sensory data that attracts many researchers is transportation especially emergency and accident services which is used as a case study in this paper. Experimental results from real-world services showed that the proposed method achieved significant improvement in the accuracy and performance in the selection process.     

定力螺栓断裂的机理分析

本文通过对螺栓定力断裂机理进行分析及计算,分别从失效形式及断裂源找出导致螺栓断裂的原因,并对影响预紧力的因素逐一分析,得出预紧力与摩擦系数的关系,对为各类螺栓定力做好技术储备,并提出防范措施。     

基于SAM-CS和SOFM的胃上皮肿瘤细胞图像识别

针对胃上皮肿瘤细胞图像(以下简称肿瘤细胞图像)黏结严重和信息冗余的特点,提出了一种将自适应观测矩阵的压缩感知(SAM-CS)和自组织特征映射(SOFM)神经网络相结合的算法。该算法将肿瘤细胞图像拉成列向量,然后利用通过自适应过程产生的观测矩阵,基于压缩感知理论对图像信息进行观测,产生线性观测向量,最后利用SOFM神经网络的学习算法对观测向量进行训练和分类,实现对肿瘤细胞图像的识别。实验表明,相比常用算法,该算法至少提高了4.2%的识别准确率和5.7%的运算速度。